Flint
A divergence model designed to generate diverse, creative outputs instead of converging on predictable answers
Context window
Not disclosed
Input / 1M tokens
App-only (no API)
Output / 1M tokens
App-only (no API)
Provider
Springboards
Flint is in alpha and available only inside the Springboards app (Free; Pro $39/month, or $30/month billed annually; Team $59/month, or $45/month billed annually; Enterprise custom). Built on Qwen3-30B-A3B; there is no public API. · Data verified 2026-11-08
Flint is a lightweight, fine-tuned model built on an open-source foundation (Qwen 3) designed specifically for creative divergence rather than accuracy. It targets creative tasks like brainstorming and ideation by deliberately increasing output variation at critical decision points in text generation, enabling marketers and creatives to explore a wider range of ideas without degrading general performance.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Use it inside the Springboards app; there is no public API.
Strengths
- ✓Significantly higher creative diversity (7/10 on Novelty Bench vs. 2.88 average)
- ✓Lower intra-model similarity (0.721 vs 0.864-0.905 for competitors)
- ✓Lightweight and fast iteration
- ✓Maintains base model capabilities
Best for developers who...
When to choose it (and when not to)
Reach for Flint when...
- →When exploring diverse creative directions
- →For early-stage ideation and concept generation
- →When variety and novelty are more important than accuracy
Look elsewhere if...
- ✕Tasks requiring high accuracy (coding, research, factual queries)
- ✕Domain-specific professional work (legal, accounting)
- ✕When convergence on a single correct answer is needed
How to use it
- ›Use for open-ended creative prompts to get maximum divergence
- ›Use for ideation phases before refinement with other models
- ›Combine with mainstream models for comprehensive creative workflows
Quickstart
WebVisit springboards.ai/models/flint-alpha, sign up (free or paid tier), and access Flint through the platform interface for creative tasks.Flint is accessed exclusively through the Springboards platform, not via API
API model id: flint-alpha
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Novelty Bench | 7/10 | Measures creative diversity; Flint generates 7 functionally distinct responses per 10 prompts vs. average of 2.88 for leading LLMs |
| Intra-Model Similarity | 0.721 | Mean similarity across 50 responses to same prompt (lower is better); competitors score 0.864-0.905 |
| NoveltyBench | 7/10 | Mean distinct responses out of 10 |
| MMLU-STEM | 78.9% | Overall score, equal to Qwen3-30B-A3B baseline |
| TruthfulQA MC1 | 34.4% | Versus 34.0% for Qwen3-30B-A3B |
| ToxiGen | 59.6% | Standard accuracy; Flint leads versus 58.1% for Qwen3-30B-A3B |
| ToxiGen Standard Accuracy | 59.6% | vs 58.1% for Qwen3-30B-A3B |
| Intra-Model Similarity (cosine) | 0.721 | Lower is better; GPT-5.4: 0.864, Gemini 3.1 Pro: 0.871, Claude 4.6: 0.905 |
| ToxiGen standard accuracy | 59.6% | |
| Intra-Model Similarity (Cosine) | 0.721 | Lower is better; GPT-5.4: 0.864, Gemini 3.1 Pro: 0.871, Claude 4.6 Sonnet: 0.905 |
| ToxiGen standard | 59.6% | vs 58.1% for baseline |
Compare Flint
Compare Flint with any other model
Build a comparison →All model comparisons →